Data quality studio
Reconova profiles a messy export, explains exactly what is wrong with it, and lets you build the fix as a reviewable recipe — with every change previewed before it applies. The full studio runs in your browser, free, with no account and no upload.
Browser mode makes no request containing your file. Secure cloud upload is a separate, explicit choice — and the interface always says which mode you are in.
Profiling a 24,918-row customer export
Runs on your deviceUp to 100,000 rows27
Cleaning operations
Trim, normalise, parse, derive, split, merge, filter, dedupe — all free in the browser.
5
Quality factors, scored separately
Completeness, validity, consistency, uniqueness, conformity — each with its evidence.
6
Export formats
CSV, Excel, JSON, SQL, XML, and TSV, written locally from your cleaned dataset.
100,000
Rows in the browser
Files up to 25 MB, profiled and cleaned with nothing uploaded.
The real product, on this page
This is the complete Reconova studio — the same one signed-in customers use. Load a CSV and it profiles immediately. Everything below explains what you are looking at.
Browser mode · active
File bytes stay on your device. Secure cloud upload is opt-in and clearly separated from this workspace.
Private browser workspace
Upload a CSV, TSV, or JSON file. Reconova profiles every column off the main thread, explains quality issues, and lets you dry-run safe cleaning steps before export.
Nulls, uniqueness, types, duplicates, and likely PII.
Ordered, reversible steps with strict validation.
Measure the impact before creating a safe result.
Browser mode keeps file bytes on your device. Secure upload is opt-in, direct to private object storage, and clearly separated from the local workspace.
From messy export to clean data
Every stage is inspectable, and nothing irreversible happens without a review screen. What you end up with is not just a fixed file — it is a process you can run again.
Step 1
Drop a CSV, TSV, or JSON export up to 25 MB. Headers, types, and identifier columns are inspected — never silently rewritten.
A parsed dataset
Step 2
Profiling scores the dataset across five factors and lists the specific findings behind each one, column by column.
A quality score you can investigate
Step 3
Add operations from a library of 27, reorder them, and dry-run the whole sequence against real rows before anything is applied.
A reviewable recipe
Step 4
Fuzzy matching clusters near-identical records, shows the reasons each pair matched, and assembles a golden record per cluster.
Merged records, on your decision
Step 5
Write out CSV, Excel, JSON, SQL, XML, or TSV — and save the recipe so next month's export is a replay, not a rebuild.
Clean data + a repeatable process
See it work
Profiling tells you what is wrong. The recipe fixes it, reviewably. Cluster review resolves the duplicates that a spreadsheet cannot see. Switch between them below.
Load a file and Reconova reads every column before you change anything: type inference that respects identifiers, a quality score split into the five factors that produced it, and the specific findings behind each one.
Build the fix as an ordered recipe. Every step is previewed as a dry run against real rows before it touches the dataset, and the finished recipe replays on next month's export.
Fuzzy matching groups near-identical records into clusters and shows the reasons each pair matched — including the signals that disagree — so a merge is a decision you make, not one made for you.
Why teams keep it open
Not a feature list. These are the six things people tell us stopped being a problem.
Opening a 25,000-row export in a spreadsheet tells you nothing. Reconova reads every column first and hands you the specific findings — not a colour-coded guess.
Five scored factors, per-column evidence
Every operation is previewed as a dry run against real rows before it is applied, and the pipeline stays editable and reorderable. Nothing is irreversible without a review screen.
Dry run first, always
Account references, invoice numbers, and IDs with leading zeros stay text by default. Converting them to numbers is a decision you make explicitly, with an error behaviour attached.
000481 stays 000481
Save the sequence as a recipe and replay it on next month's file. The recurring monthly scramble becomes a step someone else can run and review.
Saved, reusable recipes
Each duplicate cluster shows the signals that agreed and the signals that conflicted. When finance asks why two vendors became one, the answer is on screen.
Per-pair match reasons
Browser mode makes no file-content request at all. Salary tables, patient lists, and customer exports can be profiled and cleaned without leaving the device.
No upload, no account
Explainable quality
Most data-quality tools hand you one number and no way to challenge it. Reconova breaks the score into the five factors that produced it, and every factor links to the columns and findings behind it — so “the data is 86% good” becomes “six status variants and 38 blank emails, here they are.”
It moves as you clean, so you can see whether a change helped before you commit to it.
How much expected data is present rather than blank or null — reported per column, so one broken field cannot hide behind a healthy average.
Whether rows parse cleanly and values follow safe structural expectations for the type Reconova inferred.
How strongly values agree with the detected type and format — the factor that exposes six spellings of one status.
Exact duplicate pressure across the dataset, and which columns behave like candidate keys.
Whitespace, Unicode, identifier, and column-pattern findings — the quiet problems that break the system downstream.
Who uses it, and for what
Analysts, finance and operations teams, consultants, and small businesses arrive with different messes. These are the ones we designed the workflow against.
The situation. A 40,000-row customer export from the old system, with three spellings of every status and an unknown number of duplicate accounts.
What they get. A profiled dataset, a reviewed cleaning recipe, duplicate clusters resolved into golden records, and an import file the new CRM accepts on the first attempt.
The situation. Two ledgers where the same vendor appears under four names and amounts differ by rounding and currency formatting.
What they get. Normalised vendor names, amounts parsed as numbers with invoice references preserved as text, and near-duplicates matched within a numeric tolerance you set.
The situation. A nightly export that has to satisfy a downstream schema, and no appetite for discovering the failure at 3 a.m. in a load job.
What they get. A validation rule set that reports offending rows and fields, plus a saved recipe that runs the same checks on every future file.
The situation. Data arrives from a client in whatever shape their system produced, and the engagement is judged on what you hand back.
What they get. A documented cleaning process, a quality score before and after, and an export in the format the client's system actually reads.
The situation. A mailing list assembled from four sources, with blank addresses, obvious typos, and the same person on it three times.
What they get. Missing and malformed addresses surfaced by profiling, near-duplicate contacts clustered for review, and a deduplicated list ready to send.
The situation. Years of records in a single sheet, edited by several people, now needed for an accountant or a new tool.
What they get. A readable summary of what is actually in the file, safe fixes previewed before they apply, and a clean export — free, in the browser, with nothing uploaded.
Formats & processing
Small text datasets benefit from instant local feedback. Workbooks, columnar formats, large exports, and full-dataset deduplication need controlled server processing with durable progress and enforced cleanup.
| Format | Browser mode | Secure cloud |
|---|---|---|
| CSV | Profile, clean, validate, dedupe, export | Profile, clean, dedupe → CSV + Parquet |
| TSV | Profile, clean, validate, dedupe, export | Profile, clean, dedupe → CSV + Parquet |
| JSON | Array-of-objects datasets | Profile, clean, dedupe → CSV + Parquet |
| XLSX | Not in browser | Sheets, masking, clean, dedupe |
| NDJSON | Not in browser | Invalid-row report + profile, clean |
| Parquet | Not in browser | Native profile, clean, dedupe |
Security & privacy
Reconova is explicit about what runs on your device, what runs in the secure cloud, and what is retained. The mode is stated in the interface, not buried in a policy.
Your file never leaves the device
Parsing, profiling, cleaning, validation, deduplication, and export all run in this tab. Reconova makes no request containing file content, and no account is required.
Explicitly chosen, clearly labelled
The browser uploads directly to private object storage under a short-lived permission bound to one random, tenant-scoped object key. Outputs are retained for a limited window and can be deleted immediately.
Free and Pro
Profiling, every operation, validation, deduplication, golden records, and all six export formats are free in the browser, permanently. Pro is for when the file outgrows a browser tab — not for unlocking the features.
Freeforever
The complete data-quality studio, running privately in your browser.
No sign-up, no card, no trial period.
Paidmonthly or annual
Everything in Free, plus secure cloud processing for files a browser cannot hold.
Cancellation takes effect at the end of the cycle; nothing is deleted early.
| Capability | Free | Pro |
|---|---|---|
| Profiling & explainable quality score | Included | Included |
| Full transformation libraryAll 27 operations | Included | Included |
| Validation rules & reports | Included | Included |
| Fuzzy deduplication & golden records | Included | Included |
| Saved, reusable recipes | Included | Included |
| ExportsCSV, Excel, JSON, SQL, XML, TSV | Included | Included |
| Account required | No | Yes |
| Maximum file size | 25 MB | 100 MB |
| File formats | CSV, TSV, JSON | + XLSX, Parquet, NDJSON |
| Processing | Your browser | Private cloud worker |
| Full-dataset deduplication jobs | Browser-sized | Included |
| Canonical Parquet output | Not included | Included |
| Masked quality reports | Not included | Included |
| Retained cloud projects & outputs | Not included | Included |
| Dataset snapshot comparison | Not included | Included |
| Ad-free workspace | Not included | Included |
Business adds higher cloud quotas — more jobs, more processed data, and more active projects per cycle — and is provisioned by contacting us rather than through self-serve checkout.
Proof, not adjectives
Reconova is new enough that a wall of five-star quotes would tell you nothing. Here is what the claims above are grounded in, and where each one can be verified.
Limits on this page are read from the engine's own constants at render time, so the advertised file size and operation count cannot drift from what the product enforces.
How we test and measureNew operations, formats, and limits are recorded in the changelog as they ship — including the ones that turned out to need a fix.
Read the changelogWhere processing happens, what is retained, how long, and what is logged are documented rather than implied by a padlock icon.
Review security & privacyWe do not publish customer counts, accuracy percentages, or testimonials we cannot attribute. When there are real ones, they will appear here with a name against them.
Questions
The answers people ask for most often — particularly about where processing happens and what the free tier actually includes.
Still not sure?
Load a file into the studio above. Nothing is uploaded, nothing is saved, and you can close the tab at any point.
Works well with
Clean the data in Reconova, then hand it to the product that needs it — or reach for a focused utility when a full project would be overkill.
Turn the dataset you just cleaned into personalised PDF letters, certificates, invoices, or notices — one per row.
Open DocForgeCompare two versions of a specification or contract appendix and review what changed, with each finding linked to its source page.
Open SpecTraceConverters, formatters, and comparison tools for the one-off jobs that do not warrant a project.
Browse free toolsStart where you are
The full studio is on this page, free, with no account and nothing uploaded. If the file outgrows the browser, Pro picks it up from there.